NVIDIA "recruits" Groq, "recruits" the founder of TPU

NVIDIA successfully acquired the core inference technology and top talent of AI chip unicorn Groq through a "quasi-acquisition" technology licensing agreement, solidifying its absolute dominance in the field of artificial intelligence computing.

According to a report by The Wall Street Journal on Wednesday, NVIDIA and Groq have reached a non-exclusive licensing agreement to integrate Groq's AI inference technology into future products. As part of the deal, Groq founder and CEO Jonathan Ross, President Sunny Madra, and some core employees will join NVIDIA.

Earlier, media outlets cited investment firm Disruptive, claiming that NVIDIA would fully acquire Groq for $20 billion in cash. If true, this would be NVIDIA's largest acquisition in over thirty years since its founding. However, shortly after the news broke, NVIDIA denied the acquisition rumors, clearly stating that only a technology licensing agreement had been reached. Groq subsequently issued a statement confirming this, emphasizing that the company will continue to operate independently, with former CFO Simon Edwards taking over as CEO, and its cloud business GroqCloud will also maintain normal operations.

Although a comprehensive equity acquisition was not achieved, this transaction directly strengthens NVIDIA's layout in the critical computing segment of "inference." With the surge in demand for AI model applications from enterprises and consumers, the market focus is shifting from model training to inference applications. Following this news, NVIDIA's stock price, which has risen over 35% this year, showed stable performance in after-hours trading.

From "$20 Billion Acquisition" Rumors to Technology Licensing

On Wednesday, Eastern Time, rumors about NVIDIA's impending acquisition of Groq caused a stir in the market. Media outlets cited Alex Davis, CEO of Disruptive, which led Groq's financing in September, stating that NVIDIA agreed to acquire Groq for $20 billion in cash, and the transaction was progressing rapidly, including all assets of Groq except for its cloud business.

However, the situation quickly reversed. Both NVIDIA and Groq clearly stated that they reached a technology licensing agreement rather than a complete acquisition. This agreement aims to expand the adoption of high-performance, low-cost inference technology. Jonathan Ross stated on LinkedIn that he joined NVIDIA to assist in integrating or licensing the technology, while GroqCloud will continue to operate independently to provide AI processing services for software developers.

If the aforementioned $20 billion acquisition rumors were true, Groq's valuation would have soared nearly twofold compared to three months ago. In the latest financing round completed on September 17, Groq raised $750 million at a valuation of $6.9 billion. Although the acquisition ultimately did not materialize, this technology "acquisition" not only allowed NVIDIA to gain critical technology but also substantially weakened a potential competitor by absorbing core team members like Jonathan Ross.

Technical "Reconciliation" Targeting the Inference Market

The core of this transaction lies in Groq's unique chip architecture and the background of its founder. **Jonathan Ross is a veteran in the AI chip field, having studied under AI pioneer Yann LeCun and participated in the development of the famous Tensor Processing Unit (TPU) during his time at Google, which remains a strong competitor to NVIDIA's GPUs **

Groq was founded in 2016, and its developed chips are known as Language Processing Units (LPU). Unlike the general-purpose Graphics Processing Units (GPU) that NVIDIA excels at, LPUs are optimized specifically for the "inference" process—where AI models generate answers or predictions based on new data. Ross has stated that thanks to the embedded memory design, Groq's chips outperform GPUs in terms of generation speed, deployment efficiency, and energy consumption control.

Currently, as AI applications transition from the training phase to large-scale deployment, the market demand for inference chips is rising sharply. A group of startups, including Groq, is working to address energy consumption issues by redesigning hardware. NVIDIA's move aims to strengthen its weaknesses in the high-efficiency inference field by absorbing Groq's technology and talent, in response to competitive pressure from major clients like Google, Amazon, and OpenAI and Meta Platforms, which are seeking independence through self-developed chips.

The Valuation Surge and Expansion of a Capital Darling

As a star startup in the AI hardware field, Groq has experienced a rocket-like surge in valuation over the past year.

According to PitchBook estimates, Groq has raised over $3 billion in total funding. In August 2024, the company completed financing at a valuation of $2.8 billion; just one month later, its valuation doubled to $6.9 billion, with a financing amount of $750 million. This round of financing was led by Disruptive, with follow-on investments from BlackRock, Neuberger Berman, Samsung, Cisco, and 1789 Capital, where Donald Trump Jr. serves as a partner.

With strong capital support, Groq's business is growing rapidly. The company revealed that it currently supports AI applications for over 2 million developers, far exceeding the 356,000 from a year ago. Benefiting from the strong demand for AI acceleration chips, Groq has set its revenue target for this year at $500 million.

NVIDIA's Cash Reserves and Ecological Ambitions

This transaction reflects NVIDIA's strategic intention to leverage its massive cash reserves to accelerate the construction of its moat. As of the end of October last year, NVIDIA held $60.6 billion in cash and short-term investments, a significant increase from $13.3 billion at the beginning of 2023.

Even without large-scale acquisitions, NVIDIA has been intensively investing in the AI ecosystem recently. The company has invested in AI and energy infrastructure company Crusoe, model developer Cohere, and increased its stake in the upcoming AI cloud service provider CoreWeave. Additionally, NVIDIA plans to invest in OpenAI and inject $5 billion into Intel.

Under NVIDIA's strong pressure, the survival space for other AI chip startups is filled with uncertainty. Another notable chip startup, Cerebras Systems, announced over $1 billion in financing in September but suddenly withdrew its IPO application in October, highlighting the difficulties of challenging NVIDIA's dominant position in this field